CODEWITHSUNDEEP
r
Helen
Helen

Reputation: 75

Calculate cohens d for all pairs of groups in dataframe

Using the r package "effsize" I am trying to calculate cohens d between all pairs of groups in my data outputting all the pairwise d estimates as a matrix. I have provided some test data to illustrate this. I would want a matrix of d estimates for all pairs of groups 1, 2, and 3.

I am struggling to find where to start with this. I know that it could be done using loops but since my real data contains 1000 groups each with 6000 data points I think this would be slow. 

library("effsize")

test <- data.frame(
   score=c(2,3,42,1,2,3,4,5,5,6,8,2),
   group=c(1,1,1,1,2,2,2,2,3,3,3,3)
)

This would be similar functionality to what is provided for wilcox rank sum using pairwise.wilcox.test().

Upvotes: 1

Views: 799

Answers (1)

Rui Barradas
Rui Barradas

Reputation: 76402

All you have to do is to note that function combn outputs the combinations of n elements taken k at a time and can also apply a function to each resulting combination. In this case the question asks for combinations of 2 groups at a time and the function fun is applied to each one.

fun <- function(x) {
  cohen.d(x[[1]]$score, x[[2]]$score)
}

sp <- split(test, test$group)
cmb <- combn(sp, 2, fun)

cmb[, 1]
#[[1]]
#[1] "Cohen's d"
#
#[[2]]
#[1] "d"
#
#[[3]]
#[1] 0.5992954
#
#[[4]]
#    lower     upper 
#-1.169345  2.367936 
#
#[[5]]
#[1] 0.95
#
#[[6]]
#[1] medium
#Levels: negligible < small < medium < large

The code above can be written as a function that does all the work and returns a matrix.

cohen.d.pairwise.test <- function(DF, scoreCol, groupCol){
  fun <- function(x) {
    eff <- cohen.d(x[[1]][[scoreCol]], x[[2]][[scoreCol]])
    c(eff[["estimate"]],
      eff[["conf.int"]][1],
      eff[["conf.int"]][2],
      eff[["conf.level"]])
  }
  sp <- split(DF, DF[[groupCol]])
  cmb <- combn(sp, 2, fun)
  rownames(cmb) <- c("estimate", "lower", "upper", "conf.level")
  t(cmb)
}

cohen.d.pairwise.test(test, scoreCol = "score", groupCol = "group")
#       estimate     lower     upper conf.level
#[1,]  0.5992954 -1.169345 2.3679357       0.95
#[2,]  0.4732232 -1.281054 2.2275008       0.95
#[3,] -0.8795932 -2.691556 0.9323698       0.95

Upvotes: 1

Related Questions